Pre-and post-release bug prediction in an industrial setting

点击翻译按钮获取中文摘要

Bug fixing remains one of the biggest costs in creating and evolving commercial software systems. While bug fixing happens during all phases of development, bugs that are found by customers after the system is released cost orders of magnitude more money to fix than those caught early on; also, bugs in deployed software can lead to a diminished user experience and can affect the long-term reputation of both the system and the company that produces it. Moreover, when these bugs occur in a financial institution, the impact on either customers or the country can be detrimental. For these reasons, it is advantageous to create strategies to detect bugs as early as possible in development, and with minimal expense and effort on the part of the developers. In this project, we propose a two-phase approach to aid in semi-automated bug detection in a financial institution: (1) we predict which units of source code (file, class or method) are most likely to contain bugs so that developers can test and review those parts of the source code more carefully; and (2) when a bug is reported, we localize in which units this bug is likely to be fixed so that the developers can fix the bug sooner.